Custom RAG Chatbot Development for an EdTech Leader
Challenge
Years of vital decisions were trapped in messy Slack threads, costing employees hours of manual search time. The client needed a secure AI retrieval engine built in under two months on a tight budget. Key challenges included cleaning unstructured chat logs, scrubbing PII, and squeezing accurate, hallucination-free answers out of lightweight AWS LLMs.
Years of vital decisions were trapped in messy Slack threads, costing employees hours of manual search time. The client needed a secure AI retrieval engine built in under two months on a tight budget. Key challenges included cleaning unstructured chat logs, scrubbing PII, and squeezing accurate, hallucination-free answers out of lightweight AWS LLMs.
Solution
NIX built a state-driven agentic RAG solution using LangGraph and LangChain. Raw Slack logs are programmatically cleaned and converted into clean Question-Answer-Summary pairs, then validated by an automated quality check. To cut costs, vectors are stored directly inside an isolated AWS S3 bucket. Real-time user queries pass through an intent classifier to pull exact semantically matched context, with a simple Gradio UI powered by FastAPI on AWS App Runner for smooth testing.
NIX built a state-driven agentic RAG solution using LangGraph and LangChain. Raw Slack logs are programmatically cleaned and converted into clean Question-Answer-Summary pairs, then validated by an automated quality check. To cut costs, vectors are stored directly inside an isolated AWS S3 bucket. Real-time user queries pass through an intent classifier to pull exact semantically matched context, with a simple Gradio UI powered by FastAPI on AWS App Runner for smooth testing.
Results
NIX turned chaotic chat archives into an instant, searchable knowledge engine, launching a production-ready PoC in under two months. The system ran entirely within the client's AWS cloud, keeping internal data 100% private. High search accuracy and low operating costs proved the business value of internal AI, leading the client to extend NIX's contract for commercial AI product builds.
NIX turned chaotic chat archives into an instant, searchable knowledge engine, launching a production-ready PoC in under two months. The system ran entirely within the client's AWS cloud, keeping internal data 100% private. High search accuracy and low operating costs proved the business value of internal AI, leading the client to extend NIX's contract for commercial AI product builds.